MYAI168 MCP CLI

MYAI168 MCP wraps this site's chat models as an MCP (Model Context Protocol) server, so MCP-capable clients can call the models as "tools" without a browser. This page covers connecting from the command line or your own program using a developer key (dev_key) (Claude Code, mcp-remote, or custom code); the transport is Streamable HTTP (JSON-RPC 2.0). Each successful call deducts points automatically; no points are charged on any error.

If you use a web or desktop app client such as Claude.ai that only accepts a URL and uses OAuth, see the "MYAI168 MCP Web & App" page instead (no developer key needed).

Get a developer key at: https://www.myai168.com/en/ai/user/developer

Endpoint and authentication

• Endpoint: https://www.myai168.com/en/mcp/ (keep the trailing slash).
• Transport: MCP Streamable HTTP. Every request is a POST with a JSON-RPC 2.0 body.
• Auth: HTTP header Authorization: Bearer YOUR_DEV_KEY (required on every request).

Available tools

Every enabled AI module on this site is exposed as a tool (chat, image, video, voice, utilities, etc.). Query tools/list for the full set — it is returned dynamically based on what this site has enabled. All tools take the same arguments: input (required, the prompt / instruction / content) and search_internet (optional, set true to search the web first when the module supports it).

Common examples:

• claude_opus_4_8 — Anthropic Claude Opus 4.8
• chatgpt_5_5 — OpenAI GPT-5.5
• gemini_3_5_flash — Google Gemini 3.5 Flash

A tool name is the module code (dots replaced by underscores, e.g. qwen3.5_397b_a17b → qwen3_5_397b_a17b). Modules that need a file upload (image-to-prompt, PDF chat, speech-to-text, etc.) are callable via this text interface for now; which modules are available depends on this site's configuration.

Configure your MCP client

Using the Claude Code CLI (single line):

claude mcp add --transport http myai168 https://www.myai168.com/en/mcp/ --header "Authorization: Bearer YOUR_DEV_KEY"

Or in an MCP client config file (mcp-remote):

{
  "mcpServers": {
    "myai168": {
      "command": "npx",
      "args": ["mcp-remote", "https://www.myai168.com/en/mcp/", "--header", "Authorization: Bearer YOUR_DEV_KEY"]
    }
  }
}

curl examples

List the available tools (developer key required):

curl -s https://www.myai168.com/en/mcp/ -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_DEV_KEY" -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Call a model (developer key required):

curl -s https://www.myai168.com/en/mcp/ -H "Content-Type: application/json" -H "Authorization: Bearer YOUR_DEV_KEY" -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"claude_opus_4_8","arguments":{"input":"Hello, please introduce yourself."}}}'

Python example

List the available tools (developer key required):

import requests

url = "https://www.myai168.com/en/mcp/"
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer YOUR_DEV_KEY",
}
payload = {"jsonrpc": "2.0", "id": 1, "method": "tools/list"}
r = requests.post(url, headers=headers, json=payload)
for tool in r.json()["result"]["tools"]:
    print(tool["name"], "-", tool.get("description", ""))

Call a model (developer key required):

import requests

url = "https://www.myai168.com/en/mcp/"
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer YOUR_DEV_KEY",
}
payload = {
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
        "name": "claude_opus_4_8",
        "arguments": {"input": "Hello, please introduce yourself."},
    },
}
r = requests.post(url, headers=headers, json=payload)
data = r.json()
print(data["result"]["content"][0]["text"])

Response format

The response is a single JSON-RPC result (Content-Type: application/json). The model's answer is in result.content[0].text; the points charged and remaining balance are in result._meta. On a model error, result.isError is true.

Billing: each successful call deducts points based on the model's price; no points are charged on any error. If your balance is insufficient, the call returns a notice and the model is not invoked.